[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-ai":3},[4,16,26,35,44],{"id":5,"slug":6,"title":7,"excerpt":8,"category":9,"tags":10,"publishedAt":14,"readingMinutes":15},2,"notes-on-ai-learning","AI 应用开发完整学习路线","AI 这行分三条赛道，应用开发是普通开发者最该走的那条。学习顺序和判断标准，这篇说清。","AI 学习",[11,12,13],"AI","学习方法","知识管理","2026-08-08T02:00:00Z",6,{"id":17,"slug":18,"title":19,"excerpt":20,"category":9,"tags":21,"publishedAt":25,"readingMinutes":15},36,"agent-complexity-is-not-ability","Agent 智能体工程落地：为什么复杂链路反而不稳定","工具越多、链路越长，Agent 反而越不稳定。我踩坑总结：Workflow 优先，Agent 兜底。",[22,23,24],"Agent","AI 工程","系统设计","2026-07-24T02:00:00Z",{"id":27,"slug":28,"title":29,"excerpt":30,"category":9,"tags":31,"publishedAt":33,"readingMinutes":34},29,"agent-tool-call-design","Agent 工具调用工程规范：边界、拦截、容错、防失控","Agent 能不能可靠干活，一半看工具设计。轮次上限、高危拦截、容错兜底，一样都不能少。",[22,32,23],"工具调用","2026-07-10T02:00:00Z",5,{"id":36,"slug":37,"title":38,"excerpt":39,"category":9,"tags":40,"publishedAt":43,"readingMinutes":15},35,"rag-not-vector-demo","RAG 检索增强生成：完整流水线拆解（不止向量入库）","RAG 不是把文档丢进向量库就完事，预处理、分块、混合检索、重排、生成约束，缺一环都白搭。",[41,42,23],"RAG","检索","2026-06-26T02:00:00Z",{"id":45,"slug":46,"title":47,"excerpt":48,"category":9,"tags":49,"publishedAt":51,"readingMinutes":15},28,"rag-retrieval-evaluation","RAG 量化评测方案：召回率\u002F精准率人工测试集落地","改完 RAG 靠感觉说\"好像变好了\"？用测试集和召回率、精准率说话，才能真的定位问题。",[41,50,23],"评测","2026-06-12T02:00:00Z"]